Research On Methods Of Object Tracking Based On Correlation Filter

Object tracking is one of the research issues in the field of computer vision,and it has become an upsurge in the academic field and application field currently.Since the object tracking algorithm based on correlation filter has been proposed,a large number of researchers paid attention to it because of its advantages of speed and accuracy.In recent years,significant results have also been achieved.At the same time,however,it still faces many challenges.In the real environment,it is often disturbed by external factors such as background similarity,partial occlusion,scale changes and so on.Usually these factors will disturb the tracking of the algorithm and cause the target to drift.In view of the above problems,this paper proposes a correlation filter model that combines convolution features and key point tracking to improve the performance of traditional tracking algorithm based on correlation filter.In this paper,we combine convolution features with traditional tracking algorithm based on correlation filter to achieve the matching and tracking of the target.Firstly,the convolution feature of the image is extracted by using the VGG network model to initialize the correlation filter.And then,the convolution feature is used to achieve the match of the target,also it is applied to the training and update of the correlation filter model.The convolution feature of the image extracted by the shallow convolution layer of the VGG network has a lot of visual information of high-resolution and powerful characterization capabilities.It can be combined to the correlation filter model to achieve the tracking of object and can better deal with the tracking problems of the uncontrollable external factors in real scenes,such as the low discrimination between the object and background,the blurred texture,and so on.In addition,this paper proposes an algorithm that combines key point tracking with the traditional tracking algorithm based on correlation filter to improve the performance of tracking.Firstly,the key point tracking is achieved using the optical flow method to obtain the information of the displacement of the reliable key point,so that the displacement of the target can be calculated to improve the problem of tracking drift caused by the traditional correlation filter when the target is partially occluded;what's more,the information of displacement of the reliable key point obtained is used to calculate the scaling ratio of the target,so as to solve the problem of the change of the target scale.The two methods proposed in this paper are experimented on the VOT test platform.The evaluation results show that the algorithm's tracking performance under the complex environment is outstanding,which proves that the algorithms in this paper have higher accuracy and robustness when faced with low discrimination between the object and background,partial occlusion,scale change and other interference.